Update app.py

app.py

@@ -38,7 +38,6 @@ def model_inference(
         text = [text]
 
     inputs = processor(text=text, images=images, padding=True, return_tensors="pt").to("cuda")
-    print("inputs",inputs)
 
     assert decoding_strategy in [
         "Greedy",
@@ -54,8 +53,8 @@ def model_inference(
     generated_ids = model.generate(**inputs, bad_words_ids=BAD_WORDS_IDS, max_new_tokens=max_new_tokens,
                                 temperature=temperature, do_sample=do_sample, repetition_penalty=repetition_penalty,
                                 top_p=top_p),
-    generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
-    #generated_texts = processor.batch_decode(generated_ids[:, generation_args["input_ids"].size(1):], skip_special_tokens=True)
+    #generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
+    generated_texts = processor.batch_decode(generated_ids[:, inputs["input_ids"].size(1):], skip_special_tokens=True)
     print("INPUT:", text, "|OUTPUT:", generated_texts)
     return generated_texts[0]